Field of the Disclosure
The present disclosure relates to an image processing apparatus for processing images captured by a plurality of imaging units, an image processing method, and a storage medium for storing a program therefor.
Description of the Related Art
A conventional technique used in a monitoring camera system is known to capture images of an identical subject at the same time by using a plurality of cameras arranged in a monitoring area, and estimate the three-dimensional position of the subject. Japanese Patent No. 5454573 discusses a technique for estimating the three-dimensional position based on the principle of stereo viewing based on the positions and orientations of cameras and the positions of a subject on camera images. In a multi-camera system handling such three-dimensional information, it is necessary to predetermine the positions and orientations of the cameras through calibration.
Mami Noguchi and Takekazu Kato, “Calibration of A Plurality of Cameras without Synchronization by Marker Pursuit”, Information Processing Society of Japan Study Report-Computer Vision and Image Media (CVIM) 2005 discusses a method for estimating the position of a camera by detecting a calibration marker (hereinafter referred to as a marker) from a moving image by using a plurality of cameras. A marker refers to an object having a characteristic color and shape which make it easier to identify the positional coordinates on an image. For example, a ball having a color different from the surrounding color is used as a marker. The positions and orientations of cameras can be obtained by using parameters related to the positions and orientations of the cameras as unknown variables and resolving the variables with the coordinates of the marker on the image as an input. As a method for obtaining the positions and orientations of cameras based on the coordinates of markers on an image, a technique for calculating a fundamental matrix and a technique with Structure from Motion are known (refer to Richard Hartley and Andrew Zisserman “Multiple View Geometry in Computer Vision SECOND EDITION” Cambridge University Press 2003).
However, when determining the positions and orientations of cameras, the above-described conventional techniques cannot obtain the correct positions and orientations of the cameras unless a detection result is obtained for the suitable number of markers at suitable positions. Therefore, there has been a problem of degraded accuracy in estimating the positions and orientations of cameras. For example, if marker acquisition positions are not uniformly arranged, i.e., marker acquisition positions exist only in a specific portion of an image, the solution excessively adapts only to the specific portion resulting in large errors in other portions. Further, it is difficult for a user to grasp suitable marker positions and the suitable number of markers in order to preferably perform calibration.